The most reliable gains from AI in a small business rarely come from anything a customer sees. They come from the back office: supplier invoices, order confirmations, application forms, contracts, delivery notes and the correspondence that surrounds them. This is work with a clear structure, a clear right answer and, in most firms, a person who spends several hours a week moving information from one place to another.
Recent Swiss surveys suggest that companies are using AI widely but not yet systematically, and that the obstacles are less about the technology than about the data it has to work with.
Where Swiss firms are today
A UBS survey of around 2,500 Swiss companies, reported by Swissinfo in May 2026, found that about six in ten use AI. Small companies mostly use it for analysis, while large companies most often cite the automation of business processes. UBS economist Alessandro Bee summarised the picture: "Many companies are using AI, but few are doing so systematically."
Among SMEs specifically, an AXA study covered by Swissinfo in October 2025 found that translation and correspondence were the most common uses, cited by about half of firms surveyed. A follow-up reported in September 2026 recorded growth in AI for process optimisation, from 23% to 35%, and for data analysis.
In other words, most SMEs have moved past experimenting with text, but few have connected AI to the documents and systems that run their operations. That is where the next tranche of time savings sits.
What document automation actually involves
A typical back-office automation has four stages:
- Capture. A document arrives by email, upload or scan.
- Extraction. The relevant fields are read: supplier, amount, VAT, reference number, due date, line items.
- Checking. The extracted data is compared against something you already trust: the purchase order, the delivery note, the customer record, the contract.
- Routing. Matching items move on automatically. Anything that does not match goes to a person with the discrepancy highlighted.
The third stage is what separates useful automation from a faster way of making mistakes. At a Google Cloud event in Doha this autumn, the company's EMEA AI lead described exactly this pattern for invoice reconciliation: the invoice is checked against payment and supplier records before payment is approved, and people review the exceptions rather than every line.
The same logic applies well beyond finance. WSP, the engineering consultancy, describes building AI assistants around individual municipal permit types, which shortened some evaluations from months to weeks or days. A permit application, like an invoice, has known fields, known rules and a known authority who signs it off.
The real obstacle is the data
When EY surveyed 604 respondents at Swiss companies this year, the most frequently cited challenge was data quality and data silos, ahead of security, talent and budget. The UBS survey pointed the same way: alongside data protection and security, companies worried about inaccurate decisions caused by faulty algorithms or data.
For an SME this is practical, not abstract. If supplier names are spelled three ways in your accounting system, if purchase orders live in one tool and delivery notes in a shared folder, or if half your price list is in someone's head, an AI system will reproduce that confusion at speed. A short clean-up of master data often delivers more than any choice of model.
Location matters too. In the EY survey, 51% said it was business-critical that AI systems meet Swiss or European data protection requirements and that data is processed in Switzerland or the EU. For documents containing customer, employee or supplier details, that is a reasonable default under the Federal Act on Data Protection, and one you should settle before choosing a tool, not after.
A checklist for your first back-office automation
- Pick one document type. Supplier invoices, order confirmations or a single application form. Not "all incoming mail".
- Count the volume. How many arrive per month, and how long does each take to handle today?
- Identify the reference data. What does each document get checked against, and is that source accurate and current?
- Define a match. Agree the tolerance in advance: exact amounts, rounding, partial deliveries.
- Route exceptions to a named person. With the discrepancy shown, not just a flag.
- Decide where the data goes. Which systems store the extracted fields, where they are hosted, and who can access them.
- Keep an audit trail. Record what was extracted, what was matched and who approved each exception.
- Measure for a month. Compare handling time and error rate with the baseline before extending to a second document type.
The bottom line
Back-office automation is unglamorous, and that is its strength. The tasks are bounded, the results are measurable and the risks are manageable with a simple exception queue. The firms that do it well spend more time on their data and their matching rules than on the AI itself.
Not sure which of your document flows is the right starting point? The free AI Readiness Audit maps your recurring tasks and ranks what is worth automating first.
Sources
- Swissinfo, "Six out of ten Swiss companies use AI", 19 May 2026. link
- Swissinfo, "Swiss SMEs ramp up adoption of AI in workplaces", 8 October 2025. link
- Swissinfo, "AI not a job killer, say most Swiss businesses", 15 September 2026. link
- Euronews, "From answering questions to carrying out complex tasks: The next phase of AI", 28 September 2026. link
- WSP, "AI for the greater good", 6 January 2026. link
- EY Switzerland, "Artificial intelligence widely established in Swiss companies – but many are still in the early stages of scaling", 27 May 2026. link